An Adaptive Distributed Architecture

Abstract

The invention discloses an self-adaptive distributed system architecture, which can cope with the challenge of cloud service requirements for quality assurance and cost control due to changes in user requirements by self-adaptively adjusting the scale of a specified microservice instance. The invention observes the internal relationship between concurrency and performance from the perspective of distributed cluster level, has good model generalization ability, can be widely used in distributed clusters with different structures and behavior characteristics, and effectively reduces the work of IDC operation and maintenance and R&D personnel strength. In addition, the backbone algorithm of the present invention comes from the improvement based on Q Learning of the google deepmind team. The algorithm does not need to have correct input / output pairs, nor does it need to accurately correct sub-optimal behaviors. It is more focused on online planning and needs to be explored (in unknown domain) and compliance (existing knowledge).

Description

technical field [0001] The invention belongs to the technical field of Internet distributed systems, and in particular relates to an adaptive distributed system architecture. Background technique [0002] As the Internet industry has entered the cloud era for more than ten years, cloud distributed service architectures of various scales and application scenarios are silently creating value. Giants such as Google and BATJ have configured thousands of Tens of thousands of back-end service instances are distributed in various computer rooms on the earth, which undoubtedly increases the difficulty and workload of operation and maintenance. Of course, in order to solve such problems, new technologies have been emerging all the time. For example, for Double 11, Alibaba Cloud quickly concentrated the resources in the Internet computer room on popular businesses through virtualization mechanisms such as docker. There are the following issues to consider: [0003] 1. How much resou...

AI Technical Summary

Problems solved by technology

[0007] (1) At present, our distributed system does have a large amount of historical performance data saved, but most of the data are normal service data, and the reasons for a few failures are also various, such as program bugs that lead to abnormal processes, the first Three-party failures, configuration errors during the online process, and other reasons that cannot be resolved by changing the system scale; those scenarios that can be improved or avoided by changing the system scale are either unidentifiable from historical data, or are very few cases

[0008] (2) IDC operation and maintenance and research and development cannot spare a lot of time or review or filter the large amount of online performance data in a timely manner

Images